Vasilije Markovic is the co-founder and chief executive of Cognee, an open-source company building durable memory for AI agents. Its software combines graph and vector databases to give agents connected knowledge they can retrieve, revise, share, and carry between applications.
Originally from Montenegro, Markovic spent more than a decade in big-data engineering, analytics, product management, and data architecture, including leading data teams at Berlin technology companies. He also studied cognitive science and clinical psychology, influences evident in his approach to how software systems organize experience and retain useful context.
He co-founded Cognee in 2024 with Boris Arzentar, who joins him among the authors credited in the company’s open-source project. His account of Cognee’s early fundraising describes revising the pitch, contacting investors directly, and developing the product through successive iterations. In February 2026, Cognee announced a $7.5 million seed round led by Pebblebed, with participation from 42CAP and Vermilion Ventures.
Markovic’s early experiments included a CrewAI integration that gave specialized real-estate agents shared access to property information, audio, and text. At the 2025 AI Engineer World’s Fair, he applied the same architecture to hiring: agents analyzed developers’ commits, pull requests, repositories, and relationships, then wrote their assessments into a shared knowledge graph for subsequent agents to reuse.
- Structured long-term knowledge: Agent memory should preserve entities, relationships, changing facts, and supporting context. Markovic argues that retrieving a document mentioning an invoice is less useful than connecting that invoice to its customer, billing history, and previous resolutions.
- Memory that improves through use: Agents should update shared knowledge as they work, making new findings available to other agents and keeping inferred relationships connected to underlying facts. Developers can tune relevance through weighted relationships, temporal signals, and domain-specific logic; separate user graphs provide one potential privacy boundary.
- Open, portable agent memory: Cognee 1.0 supports migration between memory systems and export through the COGX interchange format, allowing accumulated knowledge to move across tools and providers.
- On-device agent memory: Cognee’s Rust-based portable memory engine extends the architecture to phones, robots, embedded hardware, and offline environments where privacy, latency, or connectivity make cloud-dependent memory impractical.
